SpamSieve learns as you train it, but training is not instant. Training SpamSieve with a message will not necessarily give it enough information to classify that message correctly based only on the words in the message. However, once you have added a message to the corpus, SpamSieve knows whether it is good or spam, even though it might not make the correct prediction based on word probabilities. This option causes SpamSieve to see if it knows whether a message is good or spam before trying to calculate its spam probability. This is especially useful when testing the setup in your mail program. If SpamSieve has seen the message before, it will always classify it correctly. You can disable this option if you want to see what SpamSieve would have predicted if it did not already know whether the message was good or spam.